GPG-HT: History-Aware Graph-Signal Policy for Reliable Routing
A new framework called GPG-HT (Generalized Policy Gradient with History-Aware Decision Transformer) has been proposed for reliable path planning in stochastic transportation networks. The method addresses the limitations of existing stochastic on-time arrival (SOTA) approaches, which rely only on current node and remaining budget, by incorporating historical node-edge-time observations. GPG-HT uses a Decision Transformer combined with generalized policy gradient optimization to capture non-Markovian spatial-temporal dependencies and history-dependent correlations in travel times. This enables context-aware decision-making for routing over graph signals, improving reliability under uncertainty.
Key facts
- GPG-HT stands for Generalized Policy Gradient with History-Aware Decision Transformer.
- The method is designed for reliable routing over graph signals.
- It addresses stochastic transportation networks with uncertain and correlated travel times.
- Existing SOTA methods depend only on current node and remaining budget.
- GPG-HT attends to historical node-edge-time observations.
- It captures non-Markovian spatial-temporal dependencies.
- The framework integrates a Decision Transformer with generalized policy gradient optimization.
- The work is published on arXiv with ID 2508.17218.
Entities
Institutions
- arXiv